Abstract

OBJECTIVES: We analyzed dietary patterns using reduced rank regression (RRR), and assessed how well the scores extracted by RRR predicted stroke in comparison to the scores produced by partial least squares and principal component regression models. METHODS: Dietary data at baseline were used to extract dietary patterns using the 3 methods, along with 4 response variables: body mass index, fibrinogen, interleukin-6, and low-density lipoprotein cholesterol. The analyses were based on 5,468 males and females aged 45-84 years who had no clinical cardiovascular disease, using data from the Multi-Ethnic Study of Atherosclerosis. RESULTS: The primary factor derived by RRR was positively associated with stroke incidence in both models. The first model was adjusted for sex and race and the second model was adjusted for the variables in model 1 as well as smoking, physical activity, family and sibling history of stroke, the use of any lipid-lowering medication, the use of any anti-hypertensive medication, hypertension, and history of myocardial infarction (model 1: hazard ratio [HR], 7.49; 95% confidence interval [CI], 1.66 to 33.69; p for trend=0.01; model 2: HR, 6.83; 95% CI, 1.51 to 30.87 for quintile 5 compared with the reference category; p for trend=0.02). CONCLUSIONS: Based primarily on RRR, we identified that a dietary pattern high in fats and oils, poultry, non-diet soda, processed meat, tomatoes, legumes, chicken, tuna and egg salad, and fried potatoes and low in dark-yellow and cruciferous vegetables may increase the incidence of ischemic stroke.